20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla

27:30You need to document, you need to feel this form, you need to do this and that and we have trouble to get the money from the insurance company so we need to document more to make sure the claims go through and so on and so on. So, you know, the other day I visited a small clinic in Paris and all the doctors looked very happy and I said, what happened, you know, it's not long. Well, the computer system is down today. We are just using papers and pen and they were so happy all of them. And so you have to realize that the state today is a very bad state where computers were bad bad news for them because they are drawing under this I mean this such a work documentation for many reasons.

28:08Doctors spend on average 49 % of their time doing this kind of I mean this creative task as opposed to caring for patients. It's bad for everyone, it's bad for them. Three out of four doctors suffer from symptoms of burnout. It's bad for patients. Three out of four suffer from burnout. Yeah, three out of four doctors have suffered from burnout symptoms and for the last year. And the main reason of this is they all mentioned the pressure of all these admin work, basically. What is that admin work? I didn't know this life. The biggest work they have to do is clinical documentation. So they have to document everything they hear, their decisions, and in big systems that we call EHR, you know, electronic has records, we're talking dinosaurs or systems.

28:53And the reason they need to do these documentation more and more, first is the financial reason to get rain boars by the insurance companies, especially in the US, you need to have like a very solid file with all that. Otherwise, the insurance will take the first opportunity, the first protects not to pay your claim. The second reason is for legal protection, because medical malpractice trials, you know, it's so expensive, your best protection is to document this in the right way. And so this is why mainly I mean, work started to grow bigger and bigger. and these EHRs are very, very old and clunky systems.

29:27I checked the other days, there was one specific technical action that took 227 clicks, mouse clicks, to be performed in the system. And I posted this on LinkedIn and many doctors answered and said, no, it actually is more than 300 clicks. And so this is a stage today. So, what we look at that, though, and the thing that strikes me there is like, okay, we could actually have recordings of meetings transcript, speech to text, and then also suggestions, meaning summary notes, whatever we want to do. Travel is its healthcare data. It's not me and you chatting about shopping where we could have preferences, normally gives a shit out, she said, March about what you want to buy from a supermarket.

30:03With healthcare data, does regulation and consumer protections get in the way of efficiency? Yeah, so at first the solution, I think AI will impact doctors, we could bring an AI assistant to every doctor to every clinician. This AI assistant will be aware of what's happening, what the patient is saying, what data we know about these patients, what the communication, what work needs to be done. Sorry, just to break it down. How does it know what the patient's saying? We were called meetings. We don't recall the meeting, but we capture the audio, we tell stories, we just capture it and then drop it.

30:35And so yeah, it gets a lot of context from the audio environment, what we call an ambient system. And the data itself from the consumer. How do we do that from wearables, how do we ingested data? So this egg assistant is connected to the EHR. So it will get a data from the existing source of thrills, which is this patient's record. Are the EHRs interoperable? Do they have the ability to move data between different applications and silos? So in theory, there is a new law in the US. In there, you know, epic spend billions to millions to try to block, but it finally was after I think 20 years of battle was passed, And so, yeah, there should be inter -rebole.

31:15Like the new model near Charles are like CRM systems today. They have an API, TC20 grade. Can I ask, I worry with healthcare about the incentive mechanism. And what I mean by that is that specifically, if we make people much more productive and we maybe remove the need for certain nurses, actually more people will be unemployed. And when we think about politicians who are allocating money for the NHS, which is obviously our healthcare system, they need nurses' votes. And so saying, hey, we're going to remove a load of low -level nurses through intelligent automation workflows, whatever that is.

31:49It goes against the political incentives by being more efficient. Does that become a problem? It's not a problem. The World Health Organization says that we are missing 18 million clinicians by 2030. So the state is not, it really put people out of the jobs as the reality is we are desperately looking for clinicians in every country in the world. So if you can make their lives easier, it's a big win for everyone. What are the biggest barriers then to AI having the impact on healthcare that I think we both believe it can? There's a many barrier. I think the mistake that many startups are doing in healthcare and what we did initially is we focused directly on the patients.

32:29It's too much too fast. You know, you cannot jump. Healthcare system is so complicated. You cannot regulation, you know, for many, many reasons, you can do that technically. We're not ready to do it. We don't have the data. We shoot first focus on clinicians, the way things are done today change your life with AI and then together We'll go to the next step. So I think to answer your question that one of the main problem is that we are trying to go too fast to patient -facing perfect system as spending a lot of time in hospitals I spend in nights in the emergency Services call center in Paris. This link to everyone.

33:00It's how is that? What do you see? So it's interesting I met a doctor in a cafe in Paris. I met a demo of Nabla on my phone and he told me, Alex, you're going to come to the emergency services call center next week all night with me. He turns out this guy where he's like the chief emergency physician in France, Patrick Poulou, invited me to spend nights over there. What I learned, spending nights over there, is it's a little bit like a call center. You have a first team he's taking calls. Overall in France, it's about 40 million calls a year. And the first responder, they don't make medical decisions.

33:33They are not doctors, they are not even nurses. The goal is to just to understand what is the address and what is the situation. What is the emergency? It's very hard to get that because people are panicked and you have lots of people calling for very small things, really close things, and you have real emergencies hidden. So the job is to sort through this very, very complicated situations when people are calling and they actually write a note, structural note with this information. They click submit, and then you have like regulation doctors who make the decision should we send the real emergency services should we send the firefighters or should we just have a doctor to this person over the line and so it's a two -step decision and a lot is done around documenting this situation and this thing you know is would be so easy to do with what we do at NABLA and you know what we know how to do with AI today the goal would be to keep the person who is answering the phone because because they are know how in terms of empathy, calming people down, understanding the context, very complicated situation again.

34:34Typically you have several people, there's screaming, nobody knows how many victims, what's happening, very, very complicated situations. And these people, I spent hours and hours listening, we'll head for them next to them. They are expert cutting through this darkness and understanding what's happening. So we want to keep these people, but they are typing with two fingers on the keyboard and missing some information. It takes a lot of time to get. And so having an AI assistant work alongside them listening to the call and documenting. I tried it, you know, it works already perfectly with what we have.

35:04It would be a perfect team. So you see an example. How many of the calls lead to ambulance sent out? Do you have that number? Is it that one in ten, one in twenty? What I saw the few nights, you know, maybe one in five results in services being sent. And in this case, three and four are firefighters who have medical doctors and so on, but they don't handle the most critical medical agencies are handled by 20 -gold Samu in France, and it's one in four. How receptive are doctors to when you talk about AI and then prove that they work with AI? And they're like, yes, bring it all the life, huh, see that before.

35:40So we've seen a change in that in the recent years, and up to three years ago, almost every doctor would tell me, okay, you don't understand our problems, go away. I'm already fighting against my EHR every day, I don't mean more stuff and it really changed recently. The core reason has nothing to do with AI is that the health systems are collapsing everywhere. You know, I know the situation with NHS here. It's worse in France, in the US, all the major hospital groups are losing money. Why there was a France? So in France we have a huge shortage of doctors and there are zones in France where you have to drive 200 kilometers and wait for three weeks just to see a GP.

36:17So this is obviously a problem. Global Ease of the problem is the same everywhere. It's access to healthcare. In France, you can get good doctor quickly if you know somebody in the system. In the US, you can get it if you have a lot of money. It looks different, but eventually it's a problem of not enough doctors. And their last doctor's never, or are we getting more unhealthy than ever, is the demanding increase or is the supply decreasing? Both since it takes 10 to 15 years to train doctors, decisions you made 20 years ago have a big impact today. So first there are less and less doctors. Many of them are aging and we go be retired soon and so the number is going to drop.

36:53And in terms of demand, everywhere in the world you have a huge rise of chronic disease like diabetes that takes a lot of medical resources, you know chronic disease. We always see as engineers, young healthy people we see that like easy things with medicine where oh I'm hurt, I go there is a quick diagnostic decision, maybe I spend two days in the hospital and then it's over, but what takes the resources in the healthcare system is not that is our aging people and chronic disease and often combined. I actually had a fascinating stat, it was like 90 % of the cost of healthcare, it's been on the last six months of life.

37:27Can I ask you, my first ever investment in fund -fight was like, what's that for doctors and nurses in the UK? They were 100 ,000 doctors and nurses in the NHS, and then it came to getting paid. And they just don't pay, and they would deliver great that 100 ,000 dollars and losses. How do you think about actually building a business being able to deliver value in a sector that just doesn't pay and doesn't have a willingness to pay? This is the biggest difficulty when you are trying to disrupt healthcare, especially for startup. You have the bear insurance or public bear. You have the providers who are providing care and you have the patients and it's a very complex relationship with three parties and typically the one who benefits from our product is not the one who is paying for it.

38:09So typically the patient will benefit, for instance, from your better SK app, but it has to be paid by the payer, by the NHS, and the provider disagrees. They can block everything. So you have to solve this three -mass problem from day one, which is very, very, very complex. And so you have to find a balance between, as a startup, if you optimize a very small process, it's easy to deploy because it follows the lines of the existing system. So, for instance, if you do a better way to take appointments, online appointments instead of calling you don't change a Husker system with that. Even that is hard to push but it's easier because you don't change the fact that there is an appointment, there is a consultation, there is a synchronous visit so you fit into the system.

38:48If you do something too ambitious it will never fit in the system. It's I think it's impossible to go to market with something too ambitious. So I think as I thought as we have to find the right level of ambition where it fits enough in the system to be adopted but it's not too much disruptive from day one otherwise he's impossible to go to market. And he pays you. So they're very hard lessons. If you are thinking about trading a headscar startup, the very, very first question I want to hear is who is paying for the product? Who is paying? What is the business model? Then you can in the startup, you know, in YC and all startup schools in the world that tell you don't start with a solution, start with a problem.

39:27But in headscar, even starting with a problem is not enough. Start with always paying and then how do you do frames or problem for this person to pay and then what is the solution for this problem eventually. If we think about this respond to the cool though and we take intelligent notes which allow them to be better and more empathetic who's going to pay for that the hospitals don't pay. So in emergency services the only potential payer is the government and so not a good good good good to market you know. Yeah this is first good to market and that's why I didn't start with that now you know we've deployed our product in the US thousands of doctors are using it every day we are well -funded now if the emergentism is in France we really want my product and the sponsor is Patrick Blueis like the god of emergency in France and Mr.

40:14Affairs look at me the eyes that say we rewant it it may be the right timing but it would be very very bad idea to start with that as a startup and maybe if you're lucky you do a pilot and then you wait and you is what I call DESPI pilot. And you go to the US and you sell the private clinics. We sell it with the US because the problem we are solving now of clinical documentation is huge in the US, big girls in Europe. And so the willingness to pay for the existing providers, and again, you need to, at entry points in the system, is very high. We found this go -to -market. We also made it possible for physicians to use our products without any approval from anyone bottom -up.

40:53They can go to nabda .com, you know, it's a web application, it's a Chrome extension, they use it tomorrow and then they love it and then we'll talk to their boss and say we like to use this in the full clinic. Can I imagine there's the US there because the bigger market. I think there's a common suggestion to a lot of AI founders today that if you want to build an AI, you've got to be in the valley. It's all about being in San Francisco. It's where the tunnel is. Paris and France has become also an AI hotspot. How do you advise founders first on whether they need to be in the valley or not to be building the best in AI.

41:24I think it used to be true, but it's not true anymore. When I built my second company with the AI, I happened to stumble about Open Add -I'm Shayer, the creator of theory, packing a lot in the valley after an event, and he helped me a lot to think about with the AI, gave me the confidence to follow what I had in mind, and owing to this confidence I was able to refuse referrals from Google and others who wanted to buy us too early. And so it had the huge impact on my company. If I were in Paris, no ad I'm share, nobody has done this before. Maybe I would have sold it to the first 10 billion. Okay, let's sell it.

41:57It's good. And so it used to be true. I think that you needed to be in the valley with having these kind of people around you. No, I think it's less true because everything is distributed. The talent is very, very distributed. It will get more distributed. And I think you should get close to your customers. Given French startups still sell to the scene? Yes, French startups. So in France, you have very good AI engineers, machine learning engineers because the education system is free, is very focused on mathematics. So we produce lots of good engineers, but we suck at growing companies. Really, we are not better at that.

42:28I don't know why. Maybe we like discipline, maybe we... Is it you suckers growing companies or you just sell to see? They also comes in and actually 50 million. It's a lot of money. If you raise one round, you'll take him 10, 15, each is co -founders, say, bowl. Exactly, say, bowl, it's enough. And so if are normal it's hard to refuse a 1 billion dollar offer if you haven't done 10 million before. But this is what Zuck did, what Google founders in SEGA and Larry did. So maybe we're not focused enough or disciplined enough and we don't have examples around us to grow this company. Hopefully it's changing.

43:03We have very good scale -ups in France now but it takes some time to. What do you think Europe needs to do to keep pace with the US in terms of AI and being an ecosystem in the last day of the day. So Europe is probably 10 years late compared to US and China, and with a new regulation, we are probably going to take 50 years more. Talk to me about that. What do you think that is? Because the regulation is so prohibitive, unpack that for me. So the new regulation is the disaster. If you look at it, people wrote this, they went too fast. Good regulation is about good timing, and involving the right people who are actually doing stuff in this domain.

43:40I think Europe being late felt like, let's be at least first in regulation. What was it about the regulation which made us so bad? Some aspects say for instance that you should be accountable, I'm simply saying that you should be accountable of the all the data you used to train your mobile, for instance. You should make sure that you have a proper license like explicit consent from everywhere it comes from and I see why it's a beautiful idea. Who would disagree with that but in practice it means one of the percent of the LMs that were trained in these last three years would be illegal in Europe.

44:13It's not connected to the reality. The intent is good, but the limited puts on how to train a model and how to operate a model makes it in practice compared to what we do today makes everything illegal. What do European startups do then? We move to the UK. Maybe for one reason Brexit? Maybe the UK. Yeah, finally Brexit is, maybe it was good. You know, I'm have to working if the regulation, if it's really like this and nobody can challenge it, startups like us may have to move. Or maybe you can keep the team in France but we have to physically train the models elsewhere, I don't know, it's a big huge danger.

44:45What would you do if you were in charge? If I put you in the hagg or in the EU and you were given the regulatory powers over the next five years of AON Europe, what would you do? First I'd wait a little bit because it's too early, we are still in a demo phase of AI, you know, LLM since not really used a lot in just beginning to be used. And as I said, you know, The regulation timing is key, it shouldn't be too late, but if you're too early and nobody knows what actually are the risks, not the risk you think because you read some science fiction books, but the real risk nobody knows yet. And so I think first I would wait maybe a year or two to learn from the field what should be regulated and why.

45:23And then I would involve people who build these models, of course, but also people who are using it, the users, the general public, which was not done so far. So they have like experts who are smart people, but there is no reality of any lens yet. What do you think about China? We spoke about Europe back in China. How do you think they've embraced AI in the next wave of AI? They have a key advantage that there is no GDPR or very few regulation internally and so the data qualities they have in the amounts, both the quality and quantity of data is incredible in China. And so for instance for healthcare data, they probably have everything about every individual in China.

45:58And if you're supported by government, I'm sure you have access for research to all these. They are lucky for that. If you think about geographies, who's the winner and who's loser in the next 10 years? China, US, Europe. They all have issues, they're issues. We talked about the advantages of China, but they are very close. They are getting more and more close. They have maybe the wrong incentives at the research level. It's hard to predict when we'd come out of China. We talked about Europe where regulation is probably the biggest problems that will keep you behind. And we are already behind that's my concern.

46:32The US is always in a very good position. The immigration laws may be something that will be a problem eventually in the US because talent is so distributed. We can learn deep learning very easily from everywhere in the world on a tablet today. It will be more and more like this. And if it's so hard to go to the US, to work for US company, eventually it will have an impact. I think I want to move to a quick fire run. I say a short statement, you give me your immediate thoughts. Does that sound okay? Okay. Okay. Do you agree that some of the biggest businesses to be built in AI will be built in services businesses helping enterprises implement AI?

47:06I disagree AI will enable a new generation of players in every industry that we use in combative and truly. What do you think is that time scale? Five years existing services companies like consulting companies are embracing AI is it will eventually remove the big data from their website and put AI and then we put an element to replace the world and say, they will take this business, I think. There was a company called Element AI a few years ago in Canada, a very ambitious company would try to do services like that and they eventually failed, they were acquired by Service No 4, so their months are raised.

47:38What was the biggest element you'd like to change about the AI community? Making more diverse, I'm really tired of talking to people like just like me. It's a cliche to say that but it would be a lot of value if we get more diverse. I have another quick anecdote. When I was at Facebook, somebody discovered that the model that was supposed to tag images that was one of the person successful to finding tennis balls in pictures. Actually, if you remove the tennis ball from the picture, it was still finding the tennis ball. So it didn't learn to find the tennis ball. It learned the context of a tennis ball, like a racket.

48:10In the second phase, it changed the color of the people on the picture. And if you had more black pixels in the hair, suddenly the model thinks, oh, it's a ping -pong ball. It was a very early days when we discovered that some models have bias. And this is bias in this case, it didn't learn to look at the ball, it learned to look at the race of the people around. And it's actually making this decision tennis versus ping pong based on the race of the people around. A man said all models have bias, is that fair? Yeah, more or less, but all of them and many models have huge bias. The guy, you know, very strong researcher, who found this thing in Facebook very, very early before everybody talked about bias.

48:47is the Mustafa CSA, is from Senegal. I don't think it's such a luck that somebody with a different angle than we have, different culture, found that a huge hole in the malls that we hadn't found before. Does AI kill traditional media? I don't think AI will kill investigative journalism, finding the right sources, the right information. So this part of media, I don't think, will be replaced by a now -to -generation part at the output, probably will be disrupted a lot by it. Was most painful as Niveland you're also pleased to have learned that Fundamentally people don't really change what may she say that I mean you can help people to grow It's true for yourself So you can correct some weaknesses you can be stronger at work where you are strong at already But I think you have to accept that some of the fundamental characteristics of yourself and the people around you won't change And if you expect them to change it leads to big fader and mistakes What do you think is the biggest misconception that people have around their eye?

49:48You hear many discussions. What do you think of that? I can't believe they're saying this again. It's that they think it's conscious. They are influenced by the form, like, oh, perfect answer. The perfect form needs a chatbot. It answers my question and influenced by the form. They make conclusion on the deep inside of the system and say, oh, it's conscious. And for me, the biggest proof of that is that in 1966, when the mother of all chatbots was released at the MIT, you know, Eliza, the very first bot. For a few months, people around the world thought that AI was solved, that there was a human, you know, sentient AI in Eliza, which is a very, very simple pattern matching system with regular expressions.

50:31But because of this, the way it was presented, people fell in the trap and I think we keep falling in this trap again and again and again for 60 years. What's the strongest belief you had which turned out to be wrong? It's very hard right here because when we change belief, we tend to forget that we have the contrary belief before and... For me like with Panda it was funny I thought like if you showed enough value you would get people to pay. If you could prove value you would get payment. Yeah it's a very good one and we had exactly the same restriction at the beginning of Lavela when I thought if patients love your healthcare product enough and you prove the health benefits of your product, then payers will pay for it.

51:11And it's not that I'm saying it's so wrong, it was obvious but probably a big belief we had that was painful to an expensive to learn. He's been the most helpful angel. Probably Jan Leuka, especially when we doubt about something or is there a confident that we are under right pass and it means a lot. What made the best VC? So I worked with Andreessen Arvitz with my previous startup. it was incredible because they never called, they never asked for anything. We didn't even have a ball meetings and every time I need something in a 20 minutes I get huge help. Maybe it's to find a real estate problem because I don't have a naffice.

51:47Maybe it's organizing a huge event and I don't want to hire a marketing team. I want to use them for one week. Or I have a strategic decision to make and they give me access to the best people in less than a day. And so the fact that you are control and you ask for help and deliver it we could go with the AI to a very nice company with only 13 employees because we are relying on 100 % of these things. It was incredible but for my first setup I did some strategic mistakes, go to market mistakes and then I wish my VCs back then would have been more involved with me and coach me and it really depend on the founder structure.

52:20A neutral VC with not here is much better than a bad VC we can have a net very negative impact on funders, I'm helping funders every day. Do you think this in Europe is good? Or do you think it's still massively behind? You've seen the US. Again, it changed a lot. So 10 years ago, VCs in Europe were all from the financial industry. There is nothing about financial industry finance in being a VC, a little bit of course, but it's mostly about entrepreneurship. So 10 years ago, all the VCs I was working with in the US were former entrepreneurs and all the VCs I knew in Europe were former bankers.

52:52In most situations, you don't need your VC to be a banker, but you need them to be an excellent entrepreneur. Tany is time, Wes Healthcare. Hopefully, ten years time, you know, every physician has their AI assistant doing a lot of stuff for them and helping them to be ten times more efficient, you know, seeing more patients, spending more time with every patient, making better decisions. I also think that I have the higher level decisions that require a big picture of the data, you know, large, large view will be taken by AI, not by people. So, for instance, in an emergency services regulation center, when you have thousands of people leading help and you have some resources, what is the state of your resources, what is the demand, how you make these decisions at the TT scale.

53:34Today is done by humans, but it's a miracle that it doesn't collapse. And sometimes it's collapsed when it would be events. 2033, where are you then, was Alison? So my dream is to build a healthcare system from scratch. Without any of all these limitations and constraints we mentioned. So controlling full stack, patient experience, provider, hospital, data -driven prevention. So my dream is to be able to build that and to do that I still need to learn a lot in healthcare and I need a few billion dollars. And this isn't I would love to end this. Thank you so much for joining me and thank you for putting up with my prior questions.

54:09Thanks, Harry. I so love that discussion. It's also so special to do it in person. If you'd like the show and want to see more from us behind the scenes, of course you can on YouTube by searching for 20VC, but before we leave you today, this episode is brought to you by TIGAS, the go -to research destination for bold investing. TIGAS curates expert insights, analysis and financial data to give you powerful perspective for your investment decisions, with lightning fast access to over 60 ,000 transcripts across 20 ,000 companies. You'll discover a wealth of unique insights to fuel your fundamental research, gain perspectives, synthesize information, model outcomes and ultimately make better decisions.

54:49All on Tegas. As a 20 minute VC listener, you can try it for free at tegas .com forward slash gross and speaking of tools we cannot live without there, Secure Frame is the leading all in one platform for automated security and privacy compliance. Secure Frame simplifies and streamlines the process of getting and staying compliant to the most rigorous global privacy and security standards like SOC2 and many more. Secure Frame's industry leading compliance automation platform paired with their in -house compliance experts and former auditors helps you get audit -ready in weeks, not months, so you can close more deals faster, delivering a suite of modern governance, risk and compliance solutions.

55:26Secureframe makes it fast and easy to achieve and maintain compliance, so you can really focus on serving your customers, scaling your business and growing your revenues, or to make your security and privacy compliance with Secureframe, SchedulerDemoData as Secureframe .com, and finally are you tired of using the same old tools and platforms in an attempt to manage your global team or so are we. Why use 16 plus different tools? When you could just use Deal, Deals are only global first HR tool with unmatched automation, customisations and integration, making it the only platform for built -in contracts, visas, background checks, payroll, equipment, benefits and beyond.

56:04So are you ready to see how Deal can scale with your global and local teams visit Deal .com ford slash 20 VC and book your demo today that's Deal .com ford slash 20 VC to book your demo today as always I so appreciate your support and I can't wait for you an incredible 20 growth on Wednesday